Training Image Operators from Samples - Research at IME-USP

Tutorial on Image Operator Learning and Applications

Abstract: High-level understanding of image contents has been receiving much attention in the last decade. Low level processing figures as a building block in this framework and it also continues to play an important role in several specific tasks such as in image filtering and colorization, medical imaging, and document image processing. The design of image operators for these tasks is usually done manually by exploiting characteristics specific to the domain of application. An alternative design approach is to use machine learning techniques to estimate the transformations. Given pairs of images consisting of a typical input and respective desired output, the goal is to estimate an operator that transforms the inputs into the desired outputs. In this tutorial we present a rigorous mathematical formulation to the framework of learning locally defined and translation invariant transformations, practical procedures and strategies to address typical machine learning related issues, application examples, and current challenges. We also include information about the code used to generate the application examples.

Bibtex:
@INPROCEEDINGS{wop-learning-tutorial, 
author={I. S. Montagner and Nina S. T. Hirata and R. Hirata Jr}, 
booktitle={29th Conference on Graphics, Patterns and Images Tutorials (SIBGRAPI-T)},
title={Image Operator Learning and Applications}, 
year={2016}
}